IBM exec calls out problem with most AI projects
For artificial intelligence (AI) projects to get off the ground, vast volumes of data are required for processing and intelligent algorithms to interpret trends, patterns and features.
In other words, AI needs learning-material, and lots of it.
Ninety-three percent of UK and US organizations consider AI to be a business priority according to a recent Vanson Bourne study commissioned by SnapLogic.
Despite the enthusiasm and ambitions for market dominance, however, companies are facing a reality check when it comes to acquiring the volumes and quality of data required.
At IBM— a company with a better view than most of the emerging technologies market— data-related struggles are a top reason the company’s clients have ceased or cancelled AI projects, according to the firm’s SVP of Cloud and Cognitive Software, Arvind Krishna.
Speaking at Wall Street Journal’s Future of Everything Festival, Krishna said that companies are finding themselves underprepared for the work and cost of acquiring and preparing that data— work comprising about 80 percent of an AI project, he told WSJ.
“[…] you run out of patience along the way, because you spend your first year just collecting and cleansing the data,” said Krishna. Companies can become impatient and disillusioned with the work, he explained, and “kind of bail on it.”
Krishna didn’t disclose the names of IBM partners that had dropped AI projects, but his words reflect a broader lack of understanding around the technology and the challenge required to bring value from it.
SnapLogic’s report found that nearly three-quarters (74 percent) of organizations have initiated an AI project in the past three years, but expectations are a long way from living up to reality.
Despite the strong levels of uptake, more than half claimed to lack the talent to execute their strategy, while around a third said budgets were holding them back. Access to useful data, as well as the right technologies and tools, was hindering progress.
With data at the heart of AI, those factors combine to make any kind of progress with projects slow— companies are going in without a real awareness of the resources it takes to succeed.
A recent report by Forrester (paywall) found that data quality was among the biggest of AI challenges. Forrester analyst Michele Goetz said producing high-quality data for use by AI software involves more than just data cleaning. Instead, data must be labelled in order to be able to provide an explanation when required about a machine’s decisions.
The data quality challenge is not to say wider industries aren’t making any progress at all. Eighty percent of companies in the financial services industry currently have AI projects in place, said SnapLogic.
IBM’s Krishna also observed that on the AI projects halted in the last five years, it was probably not far off the 50 percent estimate of IT projects at large. These setbacks, he added, are the “nature of any early technology.”